Use double rather than float.

Use double rather than float.

That means we can fit a 32-bit integer into a floating-point value
without losing precision.

Also, put in explicit casts when we convert a floating-point value to an
integral value, to suppress compiler warnings.